Background

I'm super excited about what has emerged with Pricklypear. Pricklypear is the result of a kind of long, meandering journey. I see a lot of these cactus on my hike. In the middle of the desert, they are common, they are prickly, and they bear delicious fruit. I've also been told you can use them to harvest water kefir yeast, but I've yet to confirm this.

I quit my development job and now work at a dentist office and a restaurant. But I just can't seem to stop coding. For a while I was trying to make the ultimate "CRUDmonkey" platform. Basically, I spent a lot of time with half baked designs trying to reinvent web frameworks. Then, I became totally taken with the idea of a tree-based structure editor.

My background is largely as a python CRUDmonkey, hence my interest in reinventing frameworks. I did a lot of data processing elbow grease work. I was CTO of a 13 person startup for a while. When I left that job I was pretty burnt out. I didn't think I enjoyed programming, but I still was obsessed with the idea of maybe I could make it easier.

Then I found OCaml. Functional programming did it for me, man. It was like learning to program all over again. I started learning about all the things I had basically ignored as a python guy. I learned about compilers, low level languages. I took a freelance gig doing static analysis on IBM HLASM assembly language.

The only thing I really wanted from code? I wanted to build my dream zettelkasten. I just wanted a good notes app for my bible study. First I built a bible mode in Obsidian with markdown files. But as soon as I tried to get Strong's Concordance in there, I absolutely hit the limits of what Obsidian could do. Then, I decided to switch over to elisp. LLMs were barely functional at this time, and they were really bad at lisp. This was the "copy and paste from ChatGPT" era. My emacs / org-roam zettelkasten worked amazingly. Honestly, I love emacs and I love org-roam. Absolutely beautiful pieces of software. I also love the sense of calm that I get with emacs software. I pulled up this bible app in emacs 4 years later and it just... worked. Amazingly. None of the constant churn of the JS ecosystem.

OK, so then I got into OCaml, and I was very anti LLM. They didn't work, they made you dumb, they were a scam, yada yada yada. So I built myself a bible app, all by hand, in OCaml / Dream. I had a great time with this. I felt like I probably learned more about Dream than I did OCaml, but that's how I felt when I first learned Python also.

Then I took a gig with some guys who were doing IBM mainframe LLM stuff. I started coding in TypeScript (which i had never really used), right around the time when the LLMs got really, really good. Like November of 2025? I saw a massive increase in their ability to compose and build code.

Then I went deep. I started building harnesses, orchestrators, I did the full LLM psychosis thing. I eventually settled on my Borge architecture. Without going too deeply into it, it's a DSL for defining spec and a series of CLI tools to get agents to implement, review, and build out that spec. I'm not sure it's any more efficient but it seems to work super well for me.

So then one day I had the brainwave. Can I make an interpreted lisp where the functions are stored in postgres? The agent assured me I was a genius yet again, but I had a good feeling about this one. I had the LLM draw out my Borge spec, and then started a ralph loop. I think it went for like, 25 iterations? It ran for 4 days and it crashed a couple times and I think I maxed out three different providers. It used a lot of tokens. I got an email from one of my LLM providers the next day letting me know they were tripling their costs.

When I started playing with it, I actually couldn't figure out how to use it at first. I had the agent draw up a intro guide. Then I realized that not only did it work, I could write language functions to the database, but that also those functions were working as web endpoints. I realized I had a whole framework.

It worked so well. I built a couple apps on top, it was effortless. I plugged an LLM into the system. Once I prompted it to realize Pricklypear was not a cactus in this scenario, it completely understood the harness it was in and it just... started writing functions. I couldn't believe it. Then I gave my local LLM access to the full REPL, and I realized I wouldn't ever have to build an API or CLI interface for agents ever again.

I'm delighted by this. I have wanted a Smalltalk type environment since I learned about what that was. But by incredible, happy accident, I also got my brand spanking new web framework, and I also got my tree based editor. I honestly feel, I don't need any more software. I'd be good. If I had my druthers, I would just work on this in private and read Functional Programming papers and Alonzo Church in my hammock for the next 5 years while I dogfood this thing. But, you know, "building in public" and all that.

I'm hoping some people are as thrilled and inspired by this as I am. Maybe the code is useful, or maybe just hearing about the concept is enough to inspire people. Or maybe, and this may be more likely, nobody will give a crap. I don't know where LLMs will go in the future, but I am really happy right here.